A Simple Method to Estimate the Impact of Different Radiated Emission Limits on Digital Radio Receiver Performance

Peter F. Stenumgaard, Member ,IEEE Abstract - A simple model to estimate the bit error probability (BEP) for a digital radio receiver, co-located with an undesired radiated interference source, is proposed. Co-location is defined as the area around the receiver (radius r), where an interferer has a significant impact on the BEP. The interference source is approximated to be of Gaussian type. A comparison of the emission limits given in Euronorm EN55022, Class A and B, and RE 102 in MIL-STD-461D on this radius r is presented. For Differential Phase-Shift Keying systems, assuming BEP <10-3, at SNR = 12 dB, the model delivers for Class A: r 70 m, for Class B: r 35 m and for RE102: r 10 m. Key words - Radiated Emission Standards, Undesired Sources, Bit error probability, Gaussian Approximation, Digital Radio Receiver, Co-location

I BACKGROUND
The problem of estimating the impact on digital radio receiver performance of radiated emissions arises in system design and has recently been treated in [1] and [2]. In [1], the impact on receiver sensitivity is investigated, while in [2], a methodology for estimating the performance of digital radio receivers in terms of BEP is discussed. In this paper, the findings in [2] are further developed to obtain an explicit model to estimate the impact on digital radio receiver performance of radiated emission standards. This development includes a more detailed analysis of the impact of EMIdetector used and the extension from one to several interference sources. The use of the method is examplified for Differential Phase-Shift Keying (DPSK) instead of Minimum Shift Keying (MSK), which is used in [2]. Furthermore, an error analysis of the method is performed for one of the emission limits. The interference is assumed to be undesired radiated emission from co-located electronic equipment, such as information technology devices. As radiated emission from electronic equipment, co-located to the radio receiver, can affect receiving performance, it is of great importance that this emission is considered in the early design phase of a system. Neglecting this kind of emission can cause severe problems by, for instance, taking capacity from error correcting codes in the communication system. As todays emission standards are not immediately applicable to digital radio techniques, a model is needed which can be used for system design. In most cases, the only information available to the system engineer is either limits from different radiated emission standards or results from emission radiation measurements on co-located equipment. The issue of interest is how to translate an emission level expressed in this way to bit error probability (BEP) for a digital radio receiver at a given distance from the interference source, see Fig. 1.

II THE MODELING PROBLEM

For a digital radio receiver, the bit error probability (BEP) is often used to show the impact of the receiver performance from an interfering signal. If the BEP can be estimated, this value can be compared to certain acceptable limits, depending on the kind of information transferred on the radio channel. Examples of typical limits for the BEP can be 10-3 for speech and 10-5 for data transfer. To be able to calculate BEP, the interference level at the input of the radio receiver must be known. With this information, it is possible to calculate BEP versus Signal to Noise Ratio (SNR) and separation distance between the undesired source and the radio antenna. Thus, a methodology for performing these kinds of estimations could be described in two major steps: 1) Translate the emission requirement to the interference power level that appears at the radio receiver input. 2) Determine the BEP for the radio system. Each of these steps involves different kinds of problems which must be considered in order to avoid such uncertainties that the predicted results are useless. Examples of difficulties to be handled are: - Differences in detectors used in EMI-measurements and those used in digital radio receivers. - Differences between bandwidths used in EMI-measurements and those used in radio receivers. - Extrapolation of measured results to other distances - Near field conditions during emission measurements. - Only electric field component measured. - Differences between the measurement environment of emission measurements and operation environment of the real system. - Lack of information about what kind of interference signal has given the emission measurement results. The difference in detectors used in emission measurements is in fact a more difficult problem than could be expected at a first sight, especially when we are dealing with emission from undesired sources. The type of emission from such sources can vary a lot and can be a mixture of random noise, recurrent pulses and continuous wave (CW) sinusoids. As different detectors, such as peak or quasi-peak, give different results depending on emission type [6], corrections must be made to be able to evaluate the results in a correct manner. In a real situation, it is not easy to estimate the amounts of the different emission types if only one detector is used. Comparisons between different detector responses must be made to get a good picture of the emission types involved. For instance, if the emission consists of a large amount of random noise, the peak detector gives a result which can be very difficult to handle, whereas the quasi-peak detector can be used instead. In reality, however, only emission results performed according to one standard are available, which leads to the fact that approximations must be used to determine what emission requirement should be applied on the interference source. With these difficulties in mind, a model to estimate the impact on a radio receiver, using only standard radiated emission levels, could be regarded as almost impossible. A rough model must be used, whose main purpose would be to make comparisons of the impact from different emission standards/levels to help the EMC engineer to choose a relevant emission requirement on co-located electronic equipment. A useful but rough and simple model will be proposed in the following sections.

III MODEL FOR RECEIVED INTERFERENCE POWER AT THE RADIO RECEIVER

A. Estimation of received interference power The examples of difficulties mentioned above require some assumptions to be made in order to be able to estimate the received interference power. First of all, we assume that the interference source is of an electrically small size compared to the wavelength. That is, the undesired source is small compared to the wavelength and the radio antenna is located at a distance greater than and 2 D 2 / (D = the electrical size of the undesired source). In this case, we will have conditions similar to far field, which means that antenna theory can be used to estimate the received interference power. The received interference power, PI , at the radio receiver input can then be estimated as

2 2 PI = pqGR E R (r ) , 4Z 0
where GR p q E R (r ) = E I (r ) Z0 r

(1)

wavelength[m] antenna gain of the radio antenna polarization matching factor 0 < p 1 matching factor between radio antenna impedance and load impedance, 0 < q 1 electrical field strength of the radiated interference [V/m] wave impedance for free space (= 377 ) separation distance between the undesired interference source and the radio receiver.

As the electric field is a function of the distance r from the undesired source, the rate at which it decreases with regard to r must be assumed. The problem of how to extrapolate radiated emission measurement results to different distances is a difficult problem, which is investigated, for instance in [12] and [13]. How this extrapolation should be done depends on the specific situation. For instance, theoretical calculations using a direct and a reflected wave show that for small antenna heights over good earth ( = 10 2 S/m, r = 15 ) and at frequencies above 30 MHz it is reasonable to use 1/r decay for the electric field at distances up to approximately 10 m and 1 / r 2 at distances greater than 10 m [2]. Small antenna heights mean that

r , (2) 4 where h1 =height of source [m] and h2 =height of radio antenna [m], see Fig. 2. Free space attenuation, of course, can always be used for larger distances than 10 m as a worst case approximation, as in [1]. However, as that case will be extremely conservative for small antenna heights, the suggested attenuation is likely to give more realistic results.
h1h2 << B. Impact of EMI detector used In our case, a measurement result from a standard radiated emission measurement, only the detector output from the spectrum analyzer is available. This result does not really say anything about what kind of interfering signal (noise, recurrent pulses, sinusoids etc) has been present at the antenna

input of the analyzer. As the spectrum analyzer detector output depends not only on the magnitude, but also on the type, of the interference signal the calculated interference power must be adjusted if the average power, PIav , is to be determined, for instance. In the next section, we will see that the average power is useful to determine the bit error probability. As the average power can be determined from the rms value of the amplitude, the detector output must be converted to rms value. An exact conversion to rms value is only possible if the type of interfering signal is known. For the quasi-peak detector, the conversion to rms value can be done by solving the integral equation [3] VQP Vrms where RD RC f X t ( x) VQP Vrms discharge resistance of the quasi-peak detector charge resistance of the quasi-peak detector probability density function of the detector input (output from the IF-filter) ratio between quasi-peak value and rms value. R = D RC

(x

VQP Vrms

) f X t ( x )dx ,

(3)

VQP Vrms

The brackets, , correspond to the time average value. In [3] the ratio between quasi-peak and rms value has been determined for some interference types. Equation (3) shows that without knowing anything about the interfering signal other than the envelope from the detector output, a thorough conversion to rms value is not possible. If the quasipeak detector is used, problems occur, especially if the radiated interference is dominated by recurrent pulses. An analytical approximation to the relation between quasi-peak and peak value in CISPR16 [6] is derived in [9], which gives the equation VQP / VP = 0.522 f p RD f imp RC

(
where VP fp f imp

1 VQP / VP

3/ 2

(4)

peak value pulse repetition frequency [Hz] impulse bandwidth of the analyzer.

Here, the quasi-peak value increases with the pulse repetition frequency. The relation between peak and rms value can be written as [6] VP = f imp fp

Vrms

(5)

By combining (4) and (5), the relation between quasi-peak and rms can be derived. VQP / Vrms is plotted versus f p in Fig. 3 for f imp = 120 kHz and RD / RC = 2200, which are the

values according to the CISPR16 standard. Fig. 3 shows that if the interference consists of recurrent pulses, a conversion to rms value is strongly dependent on the pulse repetition frequency. The advantage of the quasi-peak detector is that it gives an output which can be related to the variance if the input consists of noise. This relation is obtained by rewriting (3) to RD VQP = ( x VQP ) f X t ( x )dx (6) RC V
QP

and solving it for white Gaussian noise input with zero mean [14], and RD / RC = 2200. The result is obtained by numerical solution to 188 . . (7) Vrms If the peak detector has been used, problems arise if the input consists of noise, as there is a finite probability of the amplitude exceeding any predesigned value if we wait long enough [14]. Thus, a conversion from peak value to rms value for a Gaussian distribution is not possible as there is no relationship between the peak value and the statistical parameters for a noise process. One way to get some rough knowledge about the rms value in the noise case is to use crest factors [7]. Using these, a suggestion in [7] is that the ratio between peak and rms could be roughly estimated to 4. The conclusion of the discussion above is that some kind of approximation must be used if we do not know the type of interfering signal that dominates the radiated emission and if we are interested in the average power of the interference signal. VQP

IV ESTIMATION OF THE BIT ERROR PROBABILITY

To be able to make a detailed calculation of the bit error probability, the interference signal must be known in mathematical terms. The problem is that even if we can describe the interfering signal in mathematical terms, the calculation of the bit error probability is often a very complex problem except for a few interference types. In order to get a rough value of the BEP, the interfering signal can be approximated as Gaussian noise. Noise conversion refers to an old resort of treating a statistically unknown interference as if it were additional noise of the same average power [4]. This means that the average power within the spectrum analyzer bandwidth is normalized to 1 Hz, to be used as the variance of a zero mean white Gaussian process. The reason why the Gaussian approximation is a commonly used approximation is twofold. Firstly, the Gaussian distribution approximation is usually considered to give an upper bound of the BEP in several cases [4] [11]. Secondly, the Gaussian distributed interference gives fairly simple calculations. The bit error probability in the presence of Additive White Gaussian Noise (AWGN) can be calculated relatively easily for different modulation techniques, see [5] ,for instance. The BEP is a function of the signal to noise ratio (SNR), , so that Pb = Pb ( ) , (8)

where is E b / N 0 . Here, E b is energy per bit and N 0 is the power spectral density [W/Hz] of the internal noise level in the receiver. If the internal noise consists of thermal noise only, it will be equal to kT, where k is Boltzmanns constant (=138 10 23 J/K) and T is the temperature in Kelvin. . The function Pb ( ) depends on which modulation technique is used in the digital radio system. Examples of Pb ( ) are [5]

1 / 2 e 2 1 Q( / 2 ) 2 1 Q( ) 2 1 e 2 2Q( 2 ) 2Q 2 ( 2 ) where Q( x ) = 1 2

non coherent BFSK

(9)

coherent BFSK

(10)

BPSK

(11)

DPSK MSK [8],

(12) (13)

e
x

t 2

dt .

If the radio system is working under multipath fading conditions, the expressions (9) - (13) is replaced with the corresponding equations for a fading channel. If we approximate the interference as Gaussian noise within the radio receiver band, equation (8) can be used by replacing with

Eb , NI + N0

(14)

where N I is PIav normalized to a bandwidth of 1 Hz, due to the fact that the total interference is the sum of two Gaussian distributions. The interference power spectral density is NI = PIav , WM (15)

where WM is the resolution bandwidth used in the standard emission measurement. In general, the average interference power can be written as PIav = FC ( PI ) , (16)

where FC is a conversion function which depends on the detector and the dominating interference type, and PI is given by equation (1). Combining equations (1), (8), (14), (15) and (16) gives the estimated bit error probability as

 Eb . Pb = Pb ( ) = Pb 1 2 2 FC ( pqGR E R (r )) + N 0 4Z 0 WM

(17)

So, if we can convert the emission measurement result to average power, standard expressions for the bit error probability for AWGN can be used to estimate the impact on the digital radio receiver. If there are n undesired interference sources, the contribution from each of these can be handled by just adding the individual spectral densities so that N I = N In
i =1 n

(18)

The remaining question is how to convert the result from the standard emission measurement to received average power, i. e. how to determine FC . If it is possible to determine the most dominant interference type, FC can be determined for the EMI detector used. If nothing more than the standard emission measurement result is available, we do not know anything about the type of interference signal which has given the interference spectrum. If the Gaussian approximation is used, one way is to correct the result measured as if the interference signal has a Gaussian amplitude distribution, which means that the ratio between the result measured and rms value is 1.88 for the CISPR quasi-peak detector [6]. The conversion from peak value to rms value for a Gaussian distribution is, as discussed in section III, not possible as there is no relationship between the peak value and the statistical parameters for a noise process. If nothing is known about the interference type, the suggestion is to use 4 as the ratio between peak and rms value. If the approximations discussed here are used, an analysis should be made of how the introduced error in the rms conversion affects the final estimated BEP.

V EXAMPLE
A. Performance evaluation of Differential Phase-Shift Keying (DPSK) The BEP for DPSK in a Binary Symmetric Channel (BSC) with AWGN interference is 1 Pb = e , (19) 2 where is SNR defined as in the previous section. We assume that we does not have any information about the type of interfering signal, which means that the Gaussian approximation is used. If equations (17) and (19) and the suggested rms conversion for Gaussian noise are used on a digital radio receiver working with uncoded DPSK, and with an isotropic antenna with gain 0 dB, a thermal noise level of -120 dBm at 25 kHz bandwidth and working with a fixed frequency at 60 MHz, BEP versus separation distance (Fig.2) can be estimated for different values of SNR (p and q are assumed to be 1). In this example, an SNR of 12 dB has been used. Both the interference as well as the receiving antenna of the radio are located 1 m over what was above defined as good ground. The results for three different emission requirements, EN 55022 Class A and Class B and RE102 in MIL-STD-461D army material in- and outside, are shown in Fig.4 for separation distances between 10 and 200 m (As the BEP is affected by Class A equipment at distances up to approximately 200 m, it seems reasonable to include 200 m in the term co-located). An 1 / r 2 decay is used for the electric field at these distances. The emission limits for these standards are specified in Appendix A.

The maximum allowed fieldstrengths have been used, which means a worst case estimation. These maximum values are for Class A: 40 dBV/m, Class B: 30 dBV/m at a distance of 10 meters. The RE102 requirement, which is specified for 1 m distance, has been transformed to 10 m by using a 1/r decay, i.e, decreased by 20 dB which gives 4 dBV/m. At large distances the BEP will depend only on the chosen SNR. When the distance decreases, the bit error probability will increase so that, for instance, the Class A equipment causes the BEP to increase a magnitude at approximately 120 m. The result also shows the big difference between a civilian [9] and a military [10] requirement. For instance if we want lower BEP than 10 3 , which is a typical role of thumb if we want to transfer speech using no error correcting codes, the model gives minimum separation distances for Class A: 70 m and Class B: 35 m. For RE102 it only says approximately 10 m or less. These distances are for SNR= 12 dB and will decrease for larger values of SNR. It could also be interesting to see how these results depends on SNR. If the model is used for the Class B requirement for various numbers of SNR, the result in Fig. 5. is obtained. In Fig. 5, BEP has been calculated for different numbers of interference sources, by the use of equation (18), all equaling the Class B limit. Fig. 6 shows that if we increase the number of interference sources from 1 to 10, we must increase the separation distance approximately a factor 2 to have the same BEP. B. Error analysis This error analysis is performed for the Class B limit. The error analysis for the other limits can be performed in a similar way. The case analyzed here is chosen to show a situation where the approximation method gives an estimated BEP that is far below the real value, as long as the interference type is unknown. Suppose that the interfering signal in reality is dominated by a CW tone with the same frequency as the carrier frequency for the DPSK modulation. The BEP for DPSK in a mixture of CW tone and random (Gaussian) noise interference is [11] 1 2C 1/ 2 2 S 1/ 2 Pb = Q ( ) ,( ) , (20) N+R 2 N+R where t 2 + a2 Q(a , b) = I 0 (at ) exp( )tdt , (21) 2 b is the Marcum Q function, I 0 ( x ) is the 0th order modified Bessel function of the first kind and C CW tone interference power at receiving filter output S signal power at receiving filter output N thermal noise power measured in receiving filter bandwidth R random noise interference power in receiving filter bandwidth. If we assume that the interference at 60 MHz is dominated by a CW tone, with rectangular distributed phase between 0 and 2, we can use (20) to calculate the BEP. We compare this result with the BEP obtained with our proposed method, where this interference is approximated as white additive Gaussian noise with VQP Vrms = 188 . . (22)

The result is shown in Fig. 7a. In this case, the proposed method gives an estimated BEP that is far below the real value. This is due to the fact that the power of the sinewave is concentrated at one single frequency. The estimated power is derived by the use of equation (15), which spreads the power over a much greater

bandwidth (WM = 120 kHz) than the bandwidth given by the sinewave. However, as the whole power of the sinewave will go into the radio receiver, the power spectral density used in the approximation is underestimated. If we have knowledge about the interference type that has caused the emission measurement result, a better approximation can be done, see fig 7b. Here the approximation method gives an estimated BEP close to the real value. The conclusion is that one must be careful in using the proposed approximation, if the undesired radiated emission is suspected to consist of an extremely narrowband signal only. However, as undesired radiated emission in many cases is a mixture of interference types and with a bandwidth that is much greater than that for a single CW tone, for practical purposes, this drawback should not be a severe problem. Fig. 7a and 7b.

VI CONCLUSION
The proposed method makes it possible to estimate the impact of one or more interference sources on the performance (BEP) of a co-located radio receiver. This is done in two major steps: a conversion of an emission limit to an average interference power and then relating the interference noise power with the BEP. As the Gaussian noise conversion in several cases gives a worst case estimation of the BEP, the suggested model should be regarded as giving results not far from a worst case estimation of the impact on the digital radio receiver. This conclusion is based on the assumption that the conversion to rms values can be done with a good degree of accuracy. In the case that the undesired interference consists of an extremely narrowband signal only, one must be careful in using the method, as it can seriously underestimate the BEP. However, as undesired radiated in many cases is a mixture of interference types, and with a bandwidth that is much greater than that for a single CW tone, for practical purposes, the estimated performance in terms of BEP should be useful, for instance when different radiated emission limits are compared in an early design phase of a system. For lower BEP than 10 3 , at SNR= 12 dB, the model delivers for Class A: r 70 m, for Class B: r 35 m and for RE102: r 10 m, where r is the radius of the area around an interferer, where significant impact on the BEP is obtained. With the results obtained, the term co-location could be defined as the area around a radio receiver where an interference source has impact on the BEP. Thus, the term co-located is dependent on the radio receiver, the interference source and the physical environment.

ACKNOWLEDGEMENT
The author would like to thank Thomas Theiler for encouraging discussions and Jan Nilsson for manuscript review.

APPENDIX A
STANDARD EMISSION LEVELS USED IN THE CALCULATIONS
Fig. 8.

REFERENCES
[1] Virgilio P. Arafiles, Miguel J. Sanchez, Impact of Radiated Standards on Receiver Sensitivity

and the Ambient Electromagnetic Noise Environment, 1995 IEEE International EMC Symposium Records, pp 189-193. [2] Peter F. Stenumgaard, Impact of Radiated Emission Standards on Digital Radio Receiver Performance, Proceedings of EMC96 Roma, International Symposium on Electromagnetic Compatibility, pp 445 - 449. [3] J. H. Cook, Quasi-Peak-to-RMS Voltage Conversion, IEEE Transactions on Electromagnetic Compatibility, vol. EMC-21, NO. 1, February 1979. [4] Arnold S. Rosenbaum, Frederick E. Glave, An Error-Probability Upper Bound for Coherent Phase-Shift Keying with Peak-Limited Interference, IEEE Transactions on Communications, vol. COM-22, NO. 1, January 1974. [5] Simon Haykin, Digital Communications, Wiley, New York, 1988. [6] C.I.S.P.R. 16, C.I.S.P.R. Specification for radio interference measuring apparatus and measurement methods, second edition, 1987. [7] Henry W. Ott, Noise Reduction Techniques in Electronic Systems, Wiley, New York, 1988 [8] A. Svensson, The Effects of Cochannel Interference on the Error Probability of MSK-type Receivers for Continuous Phase modulation, Technical Report TR-194, University of Lund, Electrical Engineering Department, Telecommunication Theory. March 1985. [9] C.I.S.P.R. 22, Limits and methods of measurement of radio interference characteristics of information technology equipment, IEC, Geneva, 1985. [10] MIL-STD-461D. [11] J. Jay Jones, FSK and DPSK Performance in a Mixture of CW Tone and Random Noise Interference, IEEE Transactions on Communication Technology, vol. COM-18, NO.5, October 1970. [12] H. F. Garn, Problems with Radiated-Emission Testing at 3 m Distance According to CISPR11 and CISPR22, 1993 IEEE International Symposium on Electromagnetic Compatibility, pp 216-221. [13] ] H. F. Garn, Extrapolation of Radiated Emission Test Data to Distances Greater than the Distances During Test, 1990 IEEE International Symposium on Electromagnetic Compatibility, pp 456-460. [14] D. B. Geselowitz, Response of Ideal Noise Meter to Continuous Sine Wave, Recurrent Pulses, and Random Noise, IRE Transactions on Radio Frequency Interference, May 1961. [15] O. Nigol, Analysis of Radio Noise from High Voltage Lines. I-meter Response to Corona Pulses, IEEE Transactions on Power Apparatus and Systems, vol. 83, NO. 5, May 1964.